[CIG-SHORT] mesh quality
Brad Aagaard
baagaard at usgs.gov
Thu Feb 19 14:33:17 PST 2009
Tabrez-
It looks to me like it your mesh quality is still the culprit. The solution
starts blowing up in the areas with the most distorted cells. I switched to
using a small, uniform time step of 0.2 yr and it delays things blowing up
until about t = 20 years.
I think you may want to reevaluate the geometry of your domain and meshing
strategy in order to improve the mesh quality. A good rule of thumb is that
the geometry of the domain and any interior features (faults and material
boundaries) should not include scales smaller than desired resolution (cell
size).
Consider the case of a uniform mesh of 2.0 km. Including geometry that has
features significantly smaller than 2.0 km will lead to distorted cells. Your
mesh appears to have a very small scale vertical step in the top surface,
which causes cells along this interface to have very small edges. This
conflicts with the much larger desired cell size. Since you don't appear to
be interested in resolving features at such small scales, I would remove this
small scale feature from your geometry. It appears that you have also
specified a cell size larger than the thickness of the elastic layer near the
boundaries of the domain, which is also causing slightly distorted cells. In
this case you probably want to resolve the thickness of the elastic layer, so
I would reduce your cell size so that it is no larger than the thickness of
the elastic layer.
For many quasi-static problems you probably want a large domain to minimize
boundary effects and don't care about small scale features near the
boundaries. However, in close to faults you may be interested in much smaller
scale features. This leads to variable resolution meshes and juggling
constraints between matching geometry and achieving the desired mesh quality.
There are two basic strategies for refining a mesh and maintaining mesh
quality: (1) creating a smooth transition in cell size to keep cells "plump",
and (2) creating steps in cell size. For hex meshes CUBIT is limited to the
second strategy (and a factor of 3 in the length of the edges). However, for
tet meshes you can use both strategies in LaGriT and, I think, CUBIT as well.
Brad
More information about the CIG-SHORT
mailing list